The coding block processing is streamlined in cb4x4. There is no
longer a need to do special handle for sub8x8 blocks. This commit
removes the unnecessary memory allocation to the leaf nodes,
corresponding to 4x4 block size. It saves around 20% of the encoder
memory requirement.

Change-Id: I61301d6dea6c0a10ee232d166eea28578e2b3a58